didilili/ai-agents-from-zero: a Chinese MIT AI agent guide

June 13, 2026·AI Newsroom·4 min read·9 sources
open-sourcelangchainlanggraphmcpragskills+5
ai-agents-from-zero open-source repository social preview card
Image: GitHub / didilili/ai-agents-from-zero repository (MIT)

On 2026-06-13, the public repository didilili/ai-agents-from-zero — published by the Chinese Datawhale organization under the MIT license — counts 1,914 stars, 254 forks, 27 markdown chapters, two completed projects released in May 2026, and an AI interview question archive. The README presents itself as “the most systematic AI agent crash course on the web (from zero to enterprise deployment)” (README, 2026-06-13). All chapter code and both projects are Python, readable independent of the README language. The language barrier applies to prose, not samples.

What the repository covers

A single MIT syllabus spanning the 2026 production agent stack:

Repository numbers

From the GitHub REST API, 2026-06-13: 1,914 stars, 254 forks, 10 open issues, MIT license, primary language Python, size 681.8 MB, last push 2026-06-10. Repo created 2026-01-29.

The two completed projects

Both ship as separate sub-repositories, with explicit completion dates in the changelog:

The docs site is a Docsify instance (zh-CN) with sidebar, learning map, and rendered chapters. The README discloses UCloud AI cloud sponsorship — explicit, not masked.

Why it matters

1. Single-syllabus coverage of the 2026 production stack. A guide covering MCP + LangGraph + RAG + Skills + fine-tuning under MIT, in one repo, is rare. Open resources on these topics are usually scattered across 5-10 separate docs. 27 chapters in one place, with a shared glossary (全书术语表.md, 62 KB), lowers orientation costs.

2. The two projects are the real value. shopkeeper-agent and deepsearch-agents are runnable: clone, docker compose, read the Python, and you get a RAG + multi-agent + REST + WebSocket pipeline. For a team wanting a production-style reference, they’re a rare open case study, current as of May 2026.

3. Acknowledge the language barrier. The README, sidebar, 27 .md files, and interview bank are in Chinese. Python samples read fine; the prose doesn’t. Honest framing: the code reads, the prose doesn’t.

4. A Chinese industrial format. “Long tutorial + JD-aligned interview bank + completed projects” is a pattern Datawhale has industrialized for the Chinese “AI Agent / LLM Application Engineer” job profile. A specialist supplement for areas (MCP, A2A, multi-agent, eval, observability) anglophone collections treat less.

5. Supplement, not substitute. LangChain, LangGraph, and MCP each have well-maintained English docs, often more current on API details. The value of didilili is a long, organized, Chinese cross-reference that Python readers can use without translating the prose.

Practical implications

Risks and caveats

  1. Prose is Chinese, not English — verified 2026-06-13 on main. No canonical English translation yet.
  2. The syllabus aligns with the Chinese 2026 AI agent job profile. Not automatically transferable to EU/Italian market expectations.
  3. “Complete” means “27 chapters + 2 projects,” not “comprehensive.” The README calls it a “速成指南” (crash course) — a starting point, not an exhaustive reference.
  4. MIT permits forking with attribution; the article doesn’t cite third-party forks unless primary-sourced. The deepsearch-agents sub-repo doesn’t show a visible LICENSE file in GitHub metadata as of 2026-06-13.
  5. Star/fork numbers grow over time. 1,914 and 254 are 2026-06-13 snapshots.
  6. Not an installable library. No PyPI or npm releases. Clone or read, not pip install.
  7. UCloud AI cloud sponsorship is disclosed in the README — honest disclosure, flagged as transparency, not an editorial endorsement.

What to watch

Verdict

didilili/ai-agents-from-zero is a well-organized MIT guide with real Python code, two completed projects, and a structured AI interview archive. It’s not the definitive guide to AI agents and doesn’t claim to be. It’s one of the most systematic open resources on MCP + LangGraph + RAG + Skills in 2026. The value depends on tolerance for the language barrier: Python readers can use it as a cross-reference without translating the prose; readers needing only the prose will hit Chinese first. In both cases, the repository deserves to be known — a serious, MIT, Chinese, code-readable addition to the 2026 agent literature.

Sources